Matt Wolfe demonstrates a 33-minute workflow for editing talking-head shorts with Claude, reusable skills and external media tools. vidIQ sponsors the video.

Build two reusable skills

  • Analyze successful videos with vidIQ, a YouTube analytics tool connected through MCP, the standard for linking agents to tools.
  • Save separate skills for scripting and editing, including the channel’s hooks, captions, pacing and visual style.
  • Give the scripting skill a source URL and optional personal footage; review its proposed claims and shot list before recording.
  • Record your own A-roll, the main spoken footage, and let the editing skill assemble supporting B-roll.

Make cuts from sound as well as words

  • Use timestamped transcripts alongside audio waveforms so cuts preserve the ends of spoken words.
  • FFmpeg performs headless editing; Hyperframes adds graphics, Blender can create 3D inserts, and ElevenLabs supplies transcription, music or sound effects.
  • Check the rendered result for missed retakes, unnatural compression, missing music and loose pacing.
  • Save each correction into the editing skill so the next session inherits it.

The fresh-account demo needs repairs

  • An ElevenLabs transcription failure initially stops the workflow because the skill forbids alternatives.
  • Wolfe adds a fallback, repairs website permissions and resumes in a new session.
  • The first completed edit still has audio and pacing problems; a further revision takes about 36 minutes and improves the result.
  • “No code needed” describes the user interface: the agent generates Python scripts behind it.

Close the feedback loop carefully

  • Review recent videos, retention curves and comments periodically, then approve changes to the skills.
  • Audience statistics suggest experiments; the video provides no controlled evidence that these edits cause higher views.
  • Wolfe keeps human editing for long-form work and worries that cheaper shorts will displace editors and flood feeds.
  • His description mentions Premiere Pro, but the demonstrated editing pipeline uses FFmpeg without a graphical editor.

“When editing and doing initial cuts, use both the transcripts and the waveforms for editing.” — Matt Wolfe, 12:18